Source-independent Hessian-free Gauss-Newton full-waveform inversion

Wenyong Pan, Kristopher A. Innanen

Full-waveform inversion (FWI) is a powerful and promising technique for estimating the subsurface model parameters by minimizing an l-2 norm misfit function which measures the difference between the modelled data and observed data. While FWI still suffers from a lot of difficulties, one of which being the lack of source information. The estimation of source wavelet is important for successful implementation of full-waveform inversion (FWI). Many FWI algorithms estimate the source signature iteratively in the inversion process. In this paper, a source-independent method is adopted with a data calibration process. Furthermore, the gradient-based methods for FWI suffer from slow local convergence rate. A Hessian-free (HF) Gauss-Newton method is implemented in this research by solving the Newton system with a conjugate-gradient (CG) method. The Hessian-free optimization method only needs Hessian-vector products instead of constructing the Hessian matrix explicitly. In this paper, the Hessian operator in HF Gauss-Newton method is modified by combining with the source-independent strategy. We demonstrate with numerical examples that the proposed strategies can improve the convergence rate and reduce the computational burden.